Datasets:

ci-benchmark-user
/

ci-repair-bench

The dataset contains 567 CI failure instances collected from 105 real-world GitHub repositories, all written in Python.
Each instance captures a CI workflow failure, its logs, the corresponding code diff, and repository-level metadata.

Dataset Statistics

Programming language: Python
Number of repositories: 105
Number of CI instances: 567
Domain: Continuous Integration
Format: Parquet
Size category: 10M < n < 100M
Metadata language: English

Data Description

Each row in the dataset represents a CI failure scenario extracted from a GitHub repository.

Main fields

language: Programming language of the repository
id: Unique identifier of the CI instance
repo_owner: GitHub repository owner
repo_name: GitHub repository name
head_branch: Branch associated with the CI run
workflow_name: Name of the CI workflow
workflow_filename: CI workflow file name
workflow_path: Path to the workflow file
sha_fail: Commit SHA where the CI failed
sha_success: Commit SHA where the CI succeeded
workflow: Full CI workflow configuration
logs: CI execution logs capturing failure details
diff: Code diff between failing and fixed versions
changed_files: List of files modified in the fix
commit_link: Link to the failing commit
error_type: Categorized CI error type

Example error types

Code Linting
Code Formatting
Dependency Issues
Package Installation Error
Test Failure
Runtime Error
Syntax Error
Configuration Error
Environment Error
Assertion Error
Documentation / Docstring Error
Type Checking Error

Intended Use

This dataset can be used for:

CI Log Analysis
Automated Program Repair
Learning from CI logs and diffs
Benchmarking CI repair tools
Empirical software engineering research

How to Use the Dataset

Download from Hugging Face Hub

from huggingface_hub import hf_hub_download

dataset_path = hf_hub_download(
    repo_id="ci-benchmark-user/ci-repair-bench",
    filename="ci_repair_dataset.parquet",
    repo_type="dataset",
    token=config.get("HUGGINGFACE_TOKEN")
)